Chat AI

Chat AI is a conversation with an assistant that can do more than answer from memory: it searches the web, reads the files you attach, queries your databases, builds charts and dashboards, writes documents, and uses the external services you have connected. Everything it needs, it picks itself.

The screen

• Chat sessions (left pane) — your conversations, newest activity first. New chat starts a fresh thread; hover a thread to rename or delete it. The pane can be resized or collapsed.
• Conversation (centre) — messages, with the assistant's live status while it works ("Thinking…", "Searching the web…", "Using tool…") and a Jump to latest shortcut on long threads.
• Composer (bottom) — where you type. Its buttons are described below.
• Artifact panel (right, when needed) — opens automatically to show a document, chart, table or Studio page the assistant produced.

The lock badge on a thread means it is end-to-end encrypted: only your browser can read the messages (see Signing in & security).

Asking

Type and press Enter (Shift+Enter for a new line). The empty-chat suggestions are good first tries — Research a topic (web search with cited sources), Draft an email, Summarize docs, Analyze data.

The assistant keeps the whole thread in context, so follow-up questions work ("now make that a table", "only for Q3").

Attach your data

The paperclip opens a picker over My Data. You can attach, in any mix:

What	Notes
Documents (PDF, Word, PowerPoint, text, images)	Up to about 10 documents are read in full; attach more and the assistant searches across them instead, citing what it finds rather than summarising every file
Spreadsheets (CSV, Excel)	Up to 10 per turn; for questions across more tables, use a database
Databases	Table databases built by the data extractor (see Databases)
Connection tables	Tables from SQL connections you set up in My Data — the assistant queries them read-only

You can also drop files straight onto the chat: they are uploaded to My Data, trained, and attached when ready. A file still processing shows Processing…; wait for it before sending.

Attachments stay selected for the rest of the thread until you remove them (the × on the chip). The chips under a message show what was sent with it.

Skills

A skill is a saved procedure the assistant follows for a specific job — "weekly KPI report", "invoice triage" — with the tools it may use and the style it should write in. Pick one from the Skills menu to focus the conversation; choose No skill to return to the general assistant. Skills are created and managed in Workflow AI; packaged skills can be imported by administrators.

Thinking level

Thinking level (Low / Medium / High) sets how much reasoning the assistant spends on the next message. Higher is more thorough but slower; Low is fine for simple questions and quick edits.

Voice

Press the microphone to dictate. Speech is transcribed into the composer and, after 1.5 seconds of silence, sent automatically. Play audio on a reply reads it aloud.

Recipes

Recipes are reusable prompts. Save the prompt you keep typing (with rich text) and insert it later with one click. Recipes are personal; general ones can be shared by an administrator.

Reading a reply

• Sources — answers grounded in your documents cite them; click a source to open the page in the viewer. Web research shows the pages it used.
• Charts — when the assistant builds a chart it appears inline. Use Save to dashboard to pin it to a board (Dashboard).
• Files — generated documents (Word, PDF, Excel, images) appear as download cards and are kept under My Data → Artifacts.
• Studio pages — a live dashboard the assistant built opens in the artifact panel; Open in Studio takes you to the editor (Studio).
• Tables — copy a table to the clipboard with its button.

Message actions: Copy, Try again (regenerate the last reply), Play audio, and Remember this (see Memory). To rewrite part of a reply, highlight the text you want changed and describe the change.

While a long task runs (a big report, a dashboard build) you can keep the tab open or come back later — the answer is delivered to the thread even if the page was closed. A thread that is still working shows its status when you reopen it.

Memory

The assistant keeps a small long-term memory about you across all chats and workflows: preferences ("prefers tables for numbers"), durable facts ("the finance mailbox is…"), and notable events. Memories are extracted automatically after conversations; press Remember this on a reply to save something explicitly. Manage them — including Forget everything — in Profile → Memory. When a reply used memories you will see a small "using n things I know about you" note.

Dedicated agent mode

If your plan includes a Dedicated Agent, the Agent picker in the composer switches a thread from the default assistant to your own agent. Each agent keeps its own memory, so switching starts a new chat. Dedicated threads add Autopilot (hand over a whole task) and a Schedule panel for recurring work.

Tips

• One topic per thread keeps answers sharp; start a New chat when you switch subject.
• Attach the specific files you mean rather than everything — fewer, more relevant sources give better answers.
• For questions that span many spreadsheets, build a table database or connect the SQL source and attach that.
• The assistant cannot see your screen; paste the text or attach the file.

Do it with the API

Every chat action is available to integrations — see Chat AI in the API reference: POST /microservices/agent/chat/ sends a message, sessions/ manages threads, and the WebSocket ws/chat/… streams progress.